1 – Leading Machine Learning Technology
Each customer has a unique journey through a website. Recommendations AI unifies all the information available, discovers user behaviours and preferences, then forms a unique, individualised prediction for every visit.
Google describes the engine as “context-hungry”: always looking for more user events or product data to form inferences and improve prediction quality.
Recommendations AI brings significant flexibility and customisation. Three unique model types, ‘Recommended for You’, ‘Others You May Like’, and ‘Frequently Bought Together’, are suitable for different pages throughout a store. Models can also be tuned to maximise particular business objectives or apply custom filtering rules.
2 – Scalable Deployments
Recommendation engines must be able to swiftly deliver personalised predictions without increasing website loading times. Retailers know the strain of critical periods, such as seasonal sales and Christmas, with frequent surges of 3 to 10 times normal traffic.
To solve this, Recommendations AI is backed by the scale of the Google Cloud and automatically provisions itself resources on-demand. Its infrastructure is fully-managed, globally deployed, and often delivers latencies of less than 100ms.
3 – Automatic Re-Training
Initial models are only a starting point. Recommendations AI constantly tunes and re-trains models, learning from its successes and incorporating the latest products in recommendations.
Biases are a danger of black-box machine learning models. For example, sparse labels can cause popular items to be over-recommended while niche products on the long tail are underrepresented. Recommendations AI excels at handling these issues, dealing with cold-starts and adjusting for seasonality without intervention.
4 – Detailed Analytics
Once deployed, the Recommendations AI console unifies single-click maintenance tasks and comprehensive performance analytics. The dashboard details key metrics for each model and placement, allowing for tracking and comparison.
Through a system of recommendation tokens, the AI quantifies revenue gains by identifying which sales lead from engagement with the engine.
Get Started for Free
Under the managed service model, a state-of-the-art engine can be trained and deployed into production in days, ready to compare with or complement existing recommendation engines.
Google is offering $600 free credit to kick-start Recommendations AI tests, in addition to the regular $300 credit for new customers. That is enough to bring a model to production, see visible improvements in key metrics, and discover why leading e-commerce brands are making the switch.